Computer-aided quantitative stromal analysis: a comprehensive investigation of its prognostic significance in lung squamous cell carcinoma

Introduction

Lung squamous cell carcinoma (LUSC), a predominant subtype of non-small cell lung cancer (NSCLC) (1), generally derives limited clinical benefit from targeted therapies due to the scarcity of actionable molecular alterations (2). Although LUSC patients typically exhibit a higher expression of programmed cell death-ligand 1 (PD-L1) (3), multiple studies have demonstrated its suboptimal responses to immune checkpoint inhibitors (ICIs) compared to lung adenocarcinoma (4), reflecting the complexity of the tumor microenvironment (TME), molecular heterogeneity, and the multidimensional influences of immune regulatory mechanisms.

The tumor-stroma ratio (TSR), a key quantitative metric of the TME, is defined as the proportion of stromal components (including fibroblasts, immune cells, vasculature, and extracellular matrix) relative to the area of stromal tissue to the total area of the tumor tissue (including both cancer cells and stroma) (5). As a histopathological biomarker based on TME characteristics, TSR has been validated as a prognostic marker across multiple malignancies. In breast cancer, the prognostic significance of TSR varies by molecular subtype: high TSR predicts favorable outcomes in luminal subtypes but correlates with poor prognosis in triple-negative breast cancer (TNBC) (6,7). Studies in gastric cancer demonstrate strong associations between TSR and tumor aggressiveness, clinical prognosis, and peritoneal metastasis (8). Research in colorectal cancer further reveals that elevated TSR correlates with diminished disease-free survival (DFS) and overall survival (OS) (9,10). And it may serve as a valuable prognostic biomarker for intrahepatic cholangiocarcinoma (iCCA) following radical resection and demonstrates predictive potential for treatment response to adjuvant chemotherapy (11). Notably, in oral tongue squamous cell carcinoma (SCC) (12), head and neck SCC (13), and cervical SCC (14), high TSR is linked to adverse pathological features, reduced DFS, and inferior OS. In giant cell tumors of bone, TSR shows significant associations with immunosuppressive microenvironments (e.g., PD-L1 overexpression) and resistance to denosumab therapy (15). Collectively, TSR exhibits prognostic value across diverse cancers, offering potential guidance for clinical decision-making.

Similarly, existing studies have confirmed the prognostic value of the TSR in LUSC, demonstrating that patients with stroma-poor tumors exhibit superior OS and DFS compared to those with stroma-rich tumors (16). Research indicates that stroma-rich tumors are associated with poorer survival outcomes in LUSC but paradoxically correlate with favorable survival in lung adenocarcinoma (17). A recent retrospective analysis of 102 cases with unresectable locally advanced or metastatic NSCLC receiving ICIs combined with chemotherapy revealed that patients with lower TSR showed better therapeutic responses to immunotherapy compared to those with higher stromal content (18).

However, current methodologies for TSR assessment rely on manual evaluation through light microscopy, which has inherent limitations. Subjective scoring is prone to interobserver variability and field selection bias, coupled with limited reproducibility. Furthermore, conventional grouping strategies employing a fixed threshold (TSR =0.5) may overlook inter-cohort heterogeneity in TME characteristics, lacking statistically driven, objective classification criteria that compromise the sensitivity of prognostic analyses. Particularly in stroma-heterogeneous malignancies like LUSC, the low-throughput nature of manual evaluation impedes comprehensive analysis of large-scale tissue sections, potentially obscuring critical spatial heterogeneity information in key tumor regions.

In recent years, computer-aided (CA) and artificial intelligence (AI) technologies have significantly enhanced the efficiency and accuracy of TSR assessment (19,20). A study by Yang et al. demonstrated that CA quantitative TSR combined with clinical parameters can stratify high-risk stage II colon cancer patients into low-risk, intermediate-risk, and high-risk subgroups, thereby guiding adjuvant chemotherapy decisions (21). Furthermore, AI-driven TSR analysis is transitioning from research to clinical application. A machine learning model developed by the University of Oxford team, trained on 460,000 tumor nuclear images, uncovered correlations between stromal microenvironments and immunosuppression or chemotherapy resistance, offering novel insights for targeted therapeutic strategies (22).

This study introduces whole slide imaging (WSI) and a CA TSR (CA-TSR) assessment system to achieve precise quantification of tumor-stroma components. By incorporating a median-based grouping strategy, we have established a data-driven prognostic model, offering a novel approach for prognostic evaluation of patients with LUSC. We present this article in accordance with the TRIPOD reporting checklist (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-950/rc).

Methods

Research subjects and data collection

This retrospective cohort study included 220 patients with primary LUSC who underwent surgical resection at Nanjing Drum Tower Hospital between January 2015 and December 2020. Patients who received neoadjuvant therapy or had incomplete data for key variables were excluded. A total of 189 patients were ultimately included in the analysis.

We reviewed medical records to collect data on age, gender, maximum tumor diameter, tumor (T) stage, node (N) stage, metastasis (M) stage, degree of differentiation, and the status of pleural invasion, vascular invasion and neural invasion. Tumor grading and staging were performed according to the World Health Organization (WHO) classification and the 8th edition of the American Joint Committee on Cancer (AJCC) TNM classification. Follow-up for survival information was conducted until April 1, 2025. DFS was calculated from the date of surgery to the date of disease recurrence or the study cutoff, while OS was calculated from the date of surgery to the date of death or the study cutoff. The study was approved by the Ethics Committee of Nanjing Drum Tower Hospital (approval No. 2022-241-03). The study was performed in accordance with the Declaration of Helsinki and its subsequent amendments. Individual consent for this analysis was waived due to the retrospective nature.

CA-TSR assessment

The CA-TSR assessment was performed through an integrated approach that combined digital image analysis with histopathological evaluation of hematoxylin and eosin (H&E)-stained sections. An experienced senior pathologist (Q.S.) initially selected representative tissue sections that best reflected the tumor’s average TSR characteristics through microscopic examination. The selected sections were then digitally scanned at ×40 magnification using a Hamamatsu Nanozoomer^® slide scanner (Hamamatsu Photonics, Hamamatsu, Japan) to generate high-resolution images (0.23 µm/pixel). The images were subsequently processed at a resolution of 1.82 µm/pixel for analytical efficiency. CA-TSR quantification was performed using the specialized Stroma Analyzer software (Room 4, Kent, UK) developed by Danielsen et al. (23). The pathologist precisely annotated tumor parenchymal regions on the digital images, and the software automatically calculated the stromal proportion within the annotated areas to determine the final TSR values (Figure 1). This standardized protocol, which combined expert pathological judgment with objective digital analysis, ensured both clinical relevance and measurement accuracy. The median TSR score was used as the cutoff value to categorize the cases into two groups: the TSR-low (TSR-L) group, with a lower proportion of stromal components in the tumor area, and the TSR-high (TSR-H) group, with a higher proportion of stromal components.

Figure 1 Automatic analysis of tumor stroma in lung squamous cell carcinoma, with red areas marking the tumor regions, green areas representing the stroma, and blue areas indicating the background. (A) Whole-slide image of a stroma-poor case (TSR score =0.11) stained with H&E (x40). (B) Corresponding analysis image generated by the Stroma Analyzer software for Figure 1A. (C) Whole-slide image of a stroma-rich case (TSR score =0.53) stained with H&E (x40). (D) Corresponding analysis image generated by the automatic analysis software for Figure 1C. H&E, hematoxylin and eosin; TSR, tumor-stroma ratio.

Manual TSR assessment

To establish a more reliable TSR evaluation standard and validate its clinical value in predicting the prognosis of LUSC, we employed a double-blind controlled method to compare the prognostic differences between manual and CA-TSR assessments in LUSC patients. Two respiratory pathology experts (X.B. and C.X.) independently conducted manual TSR (M-TSR) evaluations on the same set of H&E-stained slides, without prior knowledge of the automated results. The evaluation process strictly adhered to a standardized protocol, utilizing a ×10 objective lens (total magnification of ×100) and ensuring that tumor cells were visible in all four quadrants of the field of view. Consistent with the automated evaluation criteria, only stromal regions in direct contact with tumor cells were assessed. At least one representative field of view was analyzed for each case. When discrepancies arose in the assessments between the two pathologists (X.B. and C.X.) (with a difference value ≥1.5-fold), a senior pathologist (Q.S.) with over 15 years of diagnostic experience was invited to review the case. A consensus assessment was reached after all three pathologists jointly reviewed the histopathological slides.

Statistical analysis

In this study, the Kolmogorov-Smirnov test was used to assess data normality. Quantitative data are presented as mean ± standard deviation, and categorical variables as frequency (percentage). Statistical methods were chosen based on data type: the Chi-squared test or Fisher’s exact probability test was used for categorical variable comparisons, and the t-test was used for normally distributed quantitative data. Survival analysis was performed using Kaplan-Meier curves with log-rank tests for intergroup comparisons. Prognostic factors were identified using Cox regression to calculate hazard ratios (HRs) and 95% confidence intervals (CIs). All analyses were conducted using SPSS 26.0, with two-tailed tests and a significance level of P<0.05.

Results

Baseline clinicopathological characteristics of the study cohort

This study enrolled 189 patients with primary LUSC (Table 1), all of whom were free of distant metastasis and underwent tumor resection surgery with lymph node dissection. The median age of the cohort was 65 years (range, 45–84 years), with a marked male predominance (186 males vs. 3 females). According to the 8th edition AJCC TNM staging system, the majority of patients were classified as T1 or T2 stage (145 cases, 76.7%), while T3–T4 stages accounted for 23.3% (44 cases). Lymph node metastasis was observed in 29.6% of cases (56 patients). Moderately differentiated tumors predominated (72.5%, 137 cases). Additionally, lymphovascular invasion, neural invasion, and pleural invasion were identified in 19% (36 cases), 14.3% (27 cases), and 18.5% (35 cases) of patients, respectively.

Association analysis between TSR score and clinicopathological features

Using a CA quantitative analysis system, we conducted TSR assessment on WSIs from surgical resection specimens of 189 primary LUSC patients (Figure 2). The median CA-TSR score was 0.34 (range, 0.11–0.9). Patients were categorized into two groups based on the median CA-TSR score: the CA-TSR-L group (stroma-poor, 98 cases) and the CA-TSR-H group (stroma-rich, 91 cases) groups based on the median value. The baseline clinicopathological characteristics of these groups are summarized in Figure 2A.

Figure 2 Clinical and pathological baseline information circular plots of the lung squamous cell carcinoma patient cohort. (A) Circular plot of clinical and pathological baseline information grouped by TSR assessed using the CA quantitative analysis system. (B) Circular plot of clinical and pathological baseline information grouped by TSR assessed through manual evaluation. CA, computer-aided; H, high; L, low; LVI, lymphovascular invasion; M, manual; N, node; NI, neural invasion; PI, pleural invasion; T, tumor; TSR, tumor-stroma ratio.

We further analyzed the correlation between CA-TSR scores and clinicopathological features (Figure 3). The results revealed that in early-stage LUSC (stage I vs. stage II), higher CA-TSR scores were associated with more advanced pathological stage (P=0.004). Conversely, in the comparison between stage II and stage III LUSC, this trend was reversed (P=0.03; Figure 3C). Compared with the CA-TSR-L group, the CA-TSR-H group exhibited higher rates of local invasion, including lymphovascular invasion (20 vs. 16 cases), neural invasion (16 vs. 11 cases), and pleural invasion (23 vs. 12 cases) (Figure 2A). Notably, CA-TSR scores were significantly correlated with pleural invasion (P=0.03, Figure 3H). Although tumor histological grade (degree of differentiation) did not reach a statistical significance with CA-TSR scores, a positive correlation trend was observed (Figure 3D), suggesting that higher CA-TSR scores may be associated with poorer tumor differentiation.

Figure 3 Correlation analysis between TSR score assessed by the computer-aided quantitative analysis system and clinicopathological features. (A) T stage. (B) N stage. (C) Pathological stage. (D) Differentiation. (E) Tumor max size. (F) LVI. (G) Neural invasion. (H) Pleural invasion. G, grade; LVI, lymphovascular invasion; N, node; T, tumor; TSR, tumor-stroma ratio.

In the manual TSR assessment, similar results were obtained. Using the median TSR score from the manual assessment as the stratification threshold (cutoff: 0.33), patients were divided into M-TSR-L (stroma-poor, 93 cases) and M-TSR-H (stroma-rich, 96 cases) groups. The distribution of baseline characteristics in these groups closely mirrored that of the CA scoring results (Figure 2B). Consistent with prior findings, higher TSR scores were associated with advanced pathological stage in early-stage disease (stage I vs. stage II: P=0.04, Figure S1), and cases with pleural invasion exhibited elevated M-TSR scores. Furthermore, comparative analysis between the two assessment methods (CA-TSR vs. M-TSR) revealed a statistically significant consistency (Spearman correlation coefficient r=0.45, P<0.001, Figure S2).

Potential prognostic value of TSR score

We conducted prognostic follow-up investigations for all LUSC patients and successfully obtained follow-up data from 157 patients. CA-TSR analysis demonstrated that patients with stromal-poor tumors (CA-TSR-L group) exhibited significantly better DFS (P<0.001, Figure 4A) and OS (P<0.001, Figure 4B) compared to those with stromal-rich tumors (CA-TSR-H group). M-TSR yielded consistent results, showing worse prognosis for stromal-rich patients with shorter DFS (P=0.03) and OS (P=0.03, Figure 4C,4D). Further Cox regression analysis revealed that pathological features, including T stage, N stage, pathological stage, vascular invasion, and pleural invasion, were significantly associated with poorer DFS and OS in univariate analysis (P<0.001, Tables S1,S2). Multivariate regression analysis demonstrated that the CA-TSR score could serve as an independent prognostic factor (HR-DFS: 2.790, P<0.001; HR-OS: 2.633, P=0.001). In contrast, M-TSR scores only showed significant prognostic value in univariate analysis (HR-DFS: 1.765, P=0.03; HR-OS: 1.770, P=0.03, Tables S3,S4). Based on the Cox regression results, we constructed nomograms for DFS and OS prediction (Figure 5). The scoring weights assigned to clinicopathological variables in the nomogram indicated that CA-TSR scores (CA-TSR-H group) contributed the most to survival risk, followed by T stage and pathological stage. The total scores calculated by summing individual variable points enable quantification of 1-, 3-, and 5-year DFS and OS probabilities for each patient.

Figure 4 Correlation between TSR and DFS as well as OS. Kaplan-Meier curves for DFS (A) and OS (B) based on TSR score assessed by the CA quantitative analysis system. Kaplan-Meier curves for DFS (C) and OS (D) based on TSR score assessed through manual evaluation. CA, computer-aided; DFS, disease-free survival; H, high; L, low; M, manual; OS, overall survival; TSR, tumor-stroma ratio.

Figure 5 Development and validation of a prognostic nomogram: a prognostic nomogram predicting the 1-, 3-, and 5-year survival probabilities of LUSC patients (TSR assessed using the CA quantitative analysis system). (A) Prognostic nomogram for DFS. (B) Prognostic nomogram for OS. CA, computer-aided; DFS, disease-free survival; H, high; L, low; LUSC, lung squamous cell carcinoma; N, node; OS, overall survival; T, tumor; TSR, tumor-stroma ratio.

Moreover, the CA-TSR score demonstrated superior predictive accuracy for patient survival outcomes. The area under the curve (AUC) values for DFS were 0.723 at 1 year, 0.674 at 3 years, and 0.658 at 5 years (Figure 6A). For OS, the corresponding AUC values were 0.666, 0.667, and 0.651 at the same timepoints (Figure 6B). Notably, the 5-year AUC values exhibited high stability with minimal fluctuation (DFS-AUC range, 0.658–0.723; OS-AUC range, 0.651–0.700) (Figure 6C,6D). In contrast, M-TSR scores showed reduced predictive performance, with DFS-related AUC values of 0.529, 0.552, and 0.572 at 1-, 3-, and 5-year intervals, respectively (Figure S3A), and OS-related AUC values of 0.488, 0.559, and 0.573 across the same follow-up periods (Figure S3B).

Figure 6 ROC curve analysis for the TSR. ROC curve analysis of the TSR in computer-aided automated analysis for DFS (A) and OS (B). Line graph of the 5-year AUC values for DFS (C) and OS (D). AUC, area under the curve; DFS, disease-free survival; FPR, false positive rate; OS, overall survival; ROC, receiver operating characteristic; TPR, true positive rate; TSR, tumor-stroma ratio.

Discussion

LUSC, a critical histological subtype of NSCLC, is characterized by distinct molecular features including low driver gene mutation rates (24), marked heterogeneity (25), and persistent lineage plasticity (26), which differentiate LUSC from adenocarcinoma and result in limited therapeutic options for targeted therapy (27) and significant heterogeneity in responses to immunotherapy (28). Although the complexity of the TME is recognized as a key contributor to treatment resistance, current research has predominantly focused on immune cell infiltration or PD-L1 expression, with insufficient attention to the dynamic regulation of stromal components and their quantitative assessment. Notably, the tumor stroma not only serves as a physical barrier for immune effector cell infiltration but also promotes tumor invasion and immune evasion through mechanisms involving activation of tumor-associated fibroblasts and modulation of the microenvironment (29,30). Previous studies assessing stromal proportion in LUSC have primarily relied on subjective interpretation (16,18), which suffers from poor reproducibility and substantial time consumption, thereby hindering precise risk stratification. In this context, we present the first application of CA quantitative analysis of stromal components in LUSC, systematically comparing its performance with that of conventional manual assessment methods and exploring its potential value in prognostic prediction.

In this study, CA automated analysis based on WSI demonstrated distinct advantages in quantifying stromal content in LUSC. Compared with conventional manual scoring, the TSR assessed by CA evaluation significantly improved prognostic accuracy (AUC-DFS: 0.723 vs. 0.529; AUC-OS: 0.666 vs. 0.488) and overcame the limitations of fixed threshold methods (e.g., TSR =0.5) in heterogeneous cohorts through a median-based stratification strategy. Notably, correlation analyses of CA-TSR scores with aggressive clinicopathological features such as pleural invasion and pathological staging, indicated that the stromal microenvironment plays a key role in local invasion in LUSC. This finding is consistent with the results of a meta-analysis conducted by Wu et al., which synthesized data from 14 studies involving 4,238 patients (31). The observed phenomenon may be attributed to activated stromal cells in stroma-rich tumors, which promote epithelial-mesenchymal transition (EMT) and the migratory capacity of cancer cells through chemokine secretion and extracellular matrix remodeling (32,33).

From the perspective of prognostic value, this study provides several key insights. First, in the multivariate Cox regression analysis, the CA-TSR score emerged as an independent risk factor (HR-DFS: 2.790; HR-OS: 2.633), highlighting its significant clinical value. This finding is consistent with the prognostic value of TSR in previous studies on LUSC and validates the reliability of stromal percentage as a stratification tool through the objectivity of CA quantitative analysis. Notably, although the M-TSR score showed prognostic significance in univariate analysis, it did not reach statistical significance in the multivariate Cox regression analysis. This highlights the inherent limitations of traditional pathological assessment, which may be related to sampling bias in field selection and subjective interpretation. Second, this study found that the relationship between CA-TSR score and pathological staging exhibited dynamic characteristics. In early-stage LUSC, higher TSR scores (TSR-H) were associated with more advanced stages, while this trend was reversed in advanced stage cases. This contradiction may reflect the dynamic evolution of the stromal microenvironment at different disease stages, and further validation through multi-center and large-sample studies is needed. Additionally, this study employed the median as the basis for TSR grouping, overcoming the adaptive limitations of traditional fixed thresholds in heterogeneous cohorts. This data-driven strategy not only avoided statistical inefficiency due to the scarcity of high stromal percentage samples (only 5 cases with TSR >0.5 in this study) but also ensured the objectivity and clinical applicability of grouping through statistical optimization, providing a more flexible and reliable analytical framework for studying stromal heterogeneity in LUSC. Finally, we compared the consistency between TSR scores assessed by CA automated system and those assessed manually, yielding a Spearman correlation coefficient of 0.45 (P<0.001). Although the consistency between CA assessment and manual TSR scoring was only moderate, CA quantitative analysis significantly reduced subjective bias through standardized algorithms. Its higher prognostic accuracy supports its central role in precision stratification. However, for hospitals lacking CA quantitative analysis facilities, manual TSR assessment could still provide an auxiliary reference for clinical decision-making by identifying stromal-rich cases and strengthening follow-up requirements for high-risk patients.

However, there are several limitations in this study. Firstly, the single-center retrospective design and male-dominated cohort may constrain the generalizability of our findings. Validation in multi-center studies with gender-balanced populations is necessary to confirm external validity. Secondly, while the TSR score quantifies stromal proportion, the functional heterogeneity of stromal components requires further clarification through integrated approaches such as single-cell sequencing or spatial transcriptomics. Finally, the exclusion of patients receiving immunotherapy precludes evaluation of the TSR score’s predictive value for immunotherapy response in LUSC, warranting future exploration in cohorts receiving immunotherapy treatment.

Conclusions

CA-TSR score demonstrates superior predictive accuracy for patient survival compared to manual TSR evaluation and is correlated with aggressive clinicopathological features like pleural invasion. CA-TSR assessment provides a highly objective, reproducible, and accurate method for prognostic risk stratification in LUSC, paving the way for its integration into clinical decision-making and future precision oncology frameworks.

Acknowledgments

We sincerely thank all patients who contributed to this research.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-950/rc

Data Sharing Statement: Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-950/dss

Peer Review File: Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-950/prf

Funding: This study was supported by the National Natural Science Foundation of China (No.82203063).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-950/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of Nanjing Drum Tower Hospital (approval No. 2022-241-03) and individual consent for this analysis was waived due to the retrospective nature.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Barta JA, Powell CA, Wisnivesky JP. Global Epidemiology of Lung Cancer. Ann Glob Health 2019;85:8. [Crossref] [PubMed]
2. Lewis WE, Hong L, Mott FE, et al. Efficacy of Targeted Inhibitors in Metastatic Lung Squamous Cell Carcinoma With EGFR or ALK Alterations. JTO Clin Res Rep 2021;2:100237. [Crossref] [PubMed]
3. Ye L, Leslie C, Jacques A, et al. Programmed death ligand-1 expression in non-small cell lung cancer in a Western Australian population and correlation with clinicopathologic features. Mod Pathol 2019;32:524-31. [Crossref] [PubMed]
4. Johnson ML, Cho BC, Luft A, et al. Durvalumab With or Without Tremelimumab in Combination With Chemotherapy as First-Line Therapy for Metastatic Non-Small-Cell Lung Cancer: The Phase III POSEIDON Study. J Clin Oncol 2023;41:1213-27. [Crossref] [PubMed]
5. van Pelt GW, Kjær-Frifeldt S, van Krieken JHJM, et al. Scoring the tumor-stroma ratio in colon cancer: procedure and recommendations. Virchows Arch 2018;473:405-12. [Crossref] [PubMed]
6. Verocq C, Noël JC, Charry M, et al. Inverse correlation between the amounts of lymphocytic infiltrate and stroma in breast carcinoma. Heliyon 2024;10:e40295. [Crossref] [PubMed]
7. Andour L, Hagenaars SC, Gregus B, et al. The prognostic value of the tumor-stroma ratio compared to tumor-infiltrating lymphocytes in triple-negative breast cancer: a review. Virchows Arch 2025;486:427-44. [Crossref] [PubMed]
8. Zhong L, Huang H, Hou D, et al. Tumor-Stroma Ratio is a Critical Indicator of Peritoneal Metastasis in Gastric Cancer. Clin Exp Gastroenterol 2025;18:11-24. [Crossref] [PubMed]
9. Shang A, Yu P, Li L, et al. Tumor stroma ratio as a clinical prognostic factor in colorectal carcinoma: A meta analysis of 7,934 patients. Oncol Lett 2025;29:190. [Crossref] [PubMed]
10. Sullivan L, Pacheco RR, Kmeid M, et al. Tumor Stroma Ratio and Its Significance in Locally Advanced Colorectal Cancer. Curr Oncol 2022;29:3232-41. [Crossref] [PubMed]
11. Zhang X, Huang CS, Huang XT, et al. Machine Learning Algorithm-Based Tumor-Stroma Ratio Can Stratify the Prognosis of Intrahepatic Cholangiocarcinoma. J Surg Res 2025;315:184-93. [Crossref] [PubMed]
12. Mansouri Z, Mohammadianpanah M, Kohandel-Shirazi M, et al. The predictive and prognostic value of tumor-stromal ratio and tumor budding in the recurrence and outcome of patients with oral tongue squamous cell carcinoma. Oral Maxillofac Surg 2024;29:17. [Crossref] [PubMed]
13. Girolami I, Damiani D, Negro R, et al. Prognostic Significance of Tumor-Stroma Ratio (TSR) in Head and Neck Squamous Cell Carcinoma: Systematic Review and Meta-Analysis. Cells 2024;13:1772. [Crossref] [PubMed]
14. Zong L, Zhang Q, Kong Y, et al. The tumor-stroma ratio is an independent predictor of survival in patients with 2018 FIGO stage IIIC squamous cell carcinoma of the cervix following primary radical surgery. Gynecol Oncol 2020;156:676-81. [Crossref] [PubMed]
15. Wu HL, Wang XB, Li J, et al. The tumor-stroma ratio in giant cell tumor of bone: associations with the immune microenvironment and responsiveness to denosumab treatment. J Orthop Surg Res 2024;19:405. [Crossref] [PubMed]
16. Wang F, Zhang Y, Li D, et al. Tumor-Stroma Ratio Is an Independent Prognostic Factor for Distant Metastasis in Squamous Cell Lung Cancer Following Resection. Can Respir J 2025;2025:9963742. [Crossref] [PubMed]
17. Zhang X, Ma H, Zhang L, et al. Predictive Role of Tumor-Stroma Ratio for Survival of Patients With Non-Small Cell Lung Cancer: A Meta-Analysis. Pathol Oncol Res 2021;27:1610021. [Crossref] [PubMed]
18. Yi L, Wen Y, Xiao M, et al. The proportion of tumour stroma predicts response to treatment of immune checkpoint inhibitor in combination with chemotherapy in patients with stage IIIB-IV non-small cell lung cancer. Histopathology 2024;85:295-309. [Crossref] [PubMed]
19. Carvalho R, Zander T, Barroso VM, et al. AI-based tumor-stroma ratio quantification algorithm: comprehensive evaluation of prognostic role in primary colorectal cancer. Virchows Arch 2025;487:799-813. [Crossref] [PubMed]
20. Baandrup L, Kjær SK, Jacobsen Ó, et al. Development of a digital algorithm for assessing tumor-stroma ratio, tumor budding and tumor infiltrating lymphocytes in vulvar squamous cell carcinomas. Ann Diagn Pathol 2025;76:152462. [Crossref] [PubMed]
21. Yang L, Chen P, Zhang L, et al. Prognostic value of nucleotyping, DNA ploidy and stroma in high-risk stage II colon cancer. Br J Cancer 2020;123:973-81. [Crossref] [PubMed]
22. Ersvaer E, Hveem TS, Vlatkovic L, et al. Prognostic value of DNA ploidy and automated assessment of stroma fraction in prostate cancer. Int J Cancer 2020;147:1228-34. [Crossref] [PubMed]
23. Danielsen HE, Hveem TS, Domingo E, et al. Prognostic markers for colorectal cancer: estimating ploidy and stroma. Ann Oncol 2018;29:616-23. [Crossref] [PubMed]
24. Zito Marino F, Bianco R, Accardo M, et al. Molecular heterogeneity in lung cancer: from mechanisms of origin to clinical implications. Int J Med Sci 2019;16:981-9. [Crossref] [PubMed]
25. Song KJ, Choi S, Kim K, et al. Proteogenomic analysis reveals non-small cell lung cancer subtypes predicting chromosome instability, and tumor microenvironment. Nat Commun 2024;15:10164. [Crossref] [PubMed]
26. Meng F, Li J, Xia Z, et al. Persistent lineage plasticity driving lung cancer development and progression. Clin Transl Med 2025;15:e70458. [Crossref] [PubMed]
27. Targeted Drugs Fall Short in Squamous Lung Cancer. Cancer Discov 2021;11:OF3. [Crossref] [PubMed]
28. Mamdani H, Matosevic S, Khalid AB, et al. Immunotherapy in Lung Cancer: Current Landscape and Future Directions. Front Immunol 2022;13:823618. [Crossref] [PubMed]
29. Butti R, Khaladkar A, Bhardwaj P, et al. Heterotypic signaling of cancer-associated fibroblasts in shaping the cancer cell drug resistance. Cancer Drug Resist 2023;6:182-204. [Crossref] [PubMed]
30. Lan X, Li W, Zhao K, et al. Revisiting the role of cancer-associated fibroblasts in tumor microenvironment. Front Immunol 2025;16:1582532. [Crossref] [PubMed]
31. Wu J, Liang C, Chen M, et al. Association between tumor-stroma ratio and prognosis in solid tumor patients: a systematic review and meta-analysis. Oncotarget 2016;7:68954-65. [Crossref] [PubMed]
32. Mai Z, Lin Y, Lin P, et al. Modulating extracellular matrix stiffness: a strategic approach to boost cancer immunotherapy. Cell Death Dis 2024;15:307. [Crossref] [PubMed]
33. Millet M, Bollmann E, Ringuette Goulet C, et al. Cancer-Associated Fibroblasts in a 3D Engineered Tissue Model Induce Tumor-like Matrix Stiffening and EMT Transition. Cancers (Basel) 2022;14:3810. [Crossref] [PubMed]